OccProphetPlus

<div align="center"> <img src="assets/occprophetplus_logo.png" alt="OccProphetPlus Logo" class="occprophetplus-logo" style="width: 75%;">

OccProphet++: Efficient and Unified End-to-End Camera-Only 4D Occupancy Forecasting

<!-- <p align="center"> <a href="https://jlchen-c.github.io/OccProphetPlus/"> <img src="https://img.shields.io/badge/Project_Page-%20?label=OccProphet%2B%2B&labelColor=f8d3d3&color=yellow"></a>&nbsp; <a href="https://arxiv.org/abs/2502.15180"> <img src="https://img.shields.io/badge/Arxiv-%20?label=OccProphet%2B%2B&labelColor=f8d3d3&color=red"></a>&nbsp; <a href="LICENSE"> <img src="https://img.shields.io/github/license/JLChen-C/OccProphetPlus?labelColor=f8d3d3"></a> </p> --> </div>

🔍 Overview

OccProphet++ is a camera-only 4D occupancy forecasting framework, offering high efficiency in both training and inference, with excellent forecasting performance.

OccProphet++ achieves a 1.8$\times$ and 4.6$\times$ speedup and 1.3% and 5.6% relative accuracy gains over OccProphet and Cam4DOcc, respectively.

OccProphet++ has the following features:

• Flexibility: OccProphet++ only relies on pure camera inputs, making it flexible and can be easily adapted to different traffic scenarios.
• High Efficiency: OccProphet++ is both training- and inference-friendly, with a unified Observer-Envisioner pipeline. Minimum 1 RTX 4090 GPU works for training and inference.
• High Performance: OccProphet++ achieves state-of-the-art performance on three real-world 4D occupancy forecasting datasets: nuScenes, Lyft-Level5 and nuScenes-Occupancy.